This package provides the user with one method that returns both a tokenizer and a model when loading HuggingFace Models, without the need for knowing which AutoModel to use.
This package utilizes the transformers library to load a tokenizer and model, without having to know the AutoModel. This means that with one command, you can easily load the tokenizer and model of a given model on HuggingFace Model Hub. This can then be easily fed into the pipeline function of transformers.
The package can be installed with the following command:
pip install hfloader
Here is a bit of code you can reference to see how to use the package.
import hfloader as hfl
huggingface_model = "cardiffnlp/twitter-roberta-base-sentiment"
tokenizer, model = hfl.load_model(huggingface_model)
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, torch_dtype=torch.bfloat16, device_map="auto")
The return of the command load_model() is going to be a tokenizer and a model, which can then be used in the pipeline method provided by transformers. It does so without the need to know which AutoModel to use, which can prove to be a hassle when trying out different models.
pip install transformers
I don't have plans to upkeep this project unless it necessitates it. I was able to achieve the goal I had set out when developing the package.